Journal Article

Forecasting recessions: can we do better on MARS?

Abstract: A number of recent articles have examined the ability of financial variables to predict recessions. In this article, Peter Sephton extends the literature by considering a non-linear, nonparametric approach to predicting the probability of recession using multivariate adaptive regression splines (MARS). The results suggest that this data-intensive approach to modeling is not a panacea for recession forecasting. Although it does well explaining the data within the sample, its out-of-sample forecasts do not improve upon the benchmark probit specification.

Keywords: Recessions; Forecasting;

Bibliographic Information

Provider: Federal Reserve Bank of St. Louis

Part of Series: Review

Publication Date: 2001

Volume: 83

Issue: Mar